On the August call reporting Nvidia's fiscal Q2 2027 quarter, Colette Kress spent one sentence rejecting a phrase and the next one supplying the number that sizes it:
We recognize the scale of this support, and we know some will call this circular financing. We see it differently. […] For context, we expect demand from the AI labs for which we expect to leverage our balance sheet to contribute toward roughly a quarter of our business next year.
"Next year" is fiscal 2028, and on the same call Nvidia guided it for the first time: revenue growing approximately 70%. That turns a vague-sounding quarter into an arithmetic one. Put the two disclosures together and the cohort whose purchases lean on Nvidia's balance sheet is worth roughly $167 billion of fiscal-2028 revenue — and standing behind it is $158.5 billion of Nvidia's own equity and guarantees.
Those two numbers are within about 5% of each other. That is the circular-financing question stated as arithmetic rather than as an accusation: about one dollar of Nvidia balance sheet for every dollar of annual revenue the supported customers are expected to produce. Neither figure is a leak or an estimate from outside. Both come from the company.
Our analysis of the print laid the disclosures out and said we would argue this separately rather than resolve it in a results piece. This is that argument.
The points
- "Roughly a quarter of our business next year" — Kress, verbatim, on the fiscal Q2 2027 call, immediately after acknowledging the circular-financing label. Disclosed.
- Fiscal 2028 revenue guided to grow approximately 70%, the first time Nvidia has guided a full year ahead. Disclosed, and explicitly described as supply-constrained rather than demand-constrained.
- Nearly $50 billion invested in the frontier AI labs, which Kress called "a small fraction of our expected free cash flow over the same period." Disclosed. For scale, the company generated $127.0 billion of free cash flow over the last four reported quarters.
- $108.5 billion of maximum gross guarantee exposure — $3.5 billion of land, power and shell guarantees for AI cloud partners, plus $105 billion entered in August with SB Energy covering roughly 4.25 gigawatts at the PORTS-Pike campus, which hosts Nvidia compute under 20-year leases to OpenAI. Disclosed.
- Over $500 billion of third-party capital being raised through new financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR. Disclosed, and it is money Nvidia is arranging rather than money Nvidia is putting up.
- Approximately 12 gigawatts of existing and planned OpenAI commitments through 2030, plus selective credit enhancement for nearly 2 gigawatts for one other lab. Disclosed.
- The qualifier most coverage dropped: Kress followed the "quarter of our business" line with "compute we ship will be consumed by investment-grade customers or those that are backed by one." Disclosed, and it materially changes what the exposure is.
What "a quarter of next year" is actually worth
Nvidia has reported two quarters of fiscal 2027 and guided a third: $81.6 billion, $96.2 billion, and $108 billion. That is $285.8 billion with one quarter still to come, and the fourth quarter is not guided.
Hold that fourth quarter flat at the third quarter's guide — deliberately conservative, since every quarter of this fiscal year has grown sequentially and the company says supply binds through fiscal 2028 — and fiscal 2027 lands near $394 billion. Grow that by the guided 70% and fiscal 2028 is roughly $670 billion. A quarter of it is about $167 billion.
The floor is worth stating too. Use only the four quarters actually reported, $303.0 billion, as though fiscal 2027 ended today: 70% growth gives $515 billion, and a quarter is $129 billion.
So the range is $129–167 billion, and the arithmetic is ours, not the company's. Nvidia gave the percentage and the growth rate; nobody has multiplied them in public. What the range does not depend on is any estimate of ours about demand — both inputs are disclosures from the same call.
Three instruments, not one
The loose version of this story treats "Nvidia is financing its customers" as a single fact. The call describes three different things, and they sit in three different places:
| Instrument | Size | Where it sits |
|---|---|---|
| Equity in frontier AI labs | ~$50B | Investments, marked through income |
| Guarantees and credit support | $108.5B | Contingent, off the revenue line |
| Third-party capital arranged | >$500B | Not Nvidia's money at all |
The equity is the smallest piece and the most visible. It is also the one already showing up in earnings: $7.8 billion of equity-securities gains ran through GAAP other income this quarter, which is why GAAP EPS of $2.46 came in above non-GAAP EPS of $2.22.
The guarantees are the largest Nvidia-borne number and the least discussed. At $108.5 billion they are 2.2× the equity, and unlike the equity they produce no gain when things go well — a guarantee is only ever worth zero or negative.
The third-party capital is the company's actual answer to the charge, and it is 10× its own commitment. If Nvidia can arrange $500 billion of institutional money against $50 billion of its own, the money going into these labs is overwhelmingly not Nvidia's. That is a real rebuttal, and it is also unproven: the platforms were announced on this call and the capital is being raised, not deployed.
The 12-gigawatt number, bounded
The tempting move is to multiply 12 gigawatts by the $40 billion per gigawatt Nvidia quotes for Vera Rubin and print $480 billion. It is worth resisting, because the company gave three per-gigawatt figures on the same call, one per generation: $18 billion with Hopper, $25 billion with Blackwell, $40 billion with Vera Rubin.
Twelve gigawatts runs through 2030 and will not all be Vera Rubin. So the honest span for OpenAI's commitment book is $216–480 billion of Nvidia revenue opportunity, and $480 billion is the ceiling — it assumes today's best generation economics hold for four more years, when the company's own series shows the figure has risen with each generation and may well keep rising. This is where most analysis quietly picks the number that suits the argument. We are not going to; anyone quoting a point estimate inside that span is inventing one.
What it does to our model
Here is the part that changes the shape of the worry. Our Nvidia model drives data-centre compute on units — rack systems shipped times price per rack — and it is bounded by a ceiling on how many racks Nvidia can ever ship in a quarter. Move the pieces and the model has an opinion about which risk is real:
| Assumption moved | Fair value | Move |
|---|---|---|
| Published base | $252.93 | — |
| Opening rack shipments cut by a quarter | $247.76 | −$5.17 |
| Sequential growth halved, 13.5% → 6.75% | $244.64 | −$8.29 |
| Shipment ceiling 65,000 → 50,000 racks | $214.83 | −$38.10 |
Losing a quarter of today's shipments costs $5.17 a share. Losing 15,000 racks of ceiling costs $38.10 — more than seven times as much. The model barely cares who the customer is. It cares almost entirely how much Nvidia can build.
That is not a quirk of our assumptions; it is the same thing Nvidia said in different words. A guide capped at 70% when "customers' forecasts point to our growth doubling" is a supply number. If the queue is longer than the line, a customer leaving it does not cost a sale — it moves the sale to whoever is next. The financing risk is therefore not really a revenue risk while supply binds, and the model has nowhere to put it because the model prices units.
Where the risk does live is the $158.5 billion, and that sits in a place no revenue model reaches: equity that can be marked down, and guarantees that can be called. Our model does not change on this disclosure, and the size of the no is the point — but it is a statement about fair value, not about safety.
What to watch
- Whether the fourth quarter of fiscal 2027 gets guided, which converts the $129–167 billion range into a single number.
- Whether the $500 billion of third-party capital actually closes. Announced platforms are not raised funds, and the 10× ratio that answers the circular-financing charge depends entirely on them landing.
- Whether guarantee exposure keeps growing. It went from $3.5 billion to $108.5 billion on one August agreement. A second one of that size would put the guarantees above a year of the revenue they support.
- The identity and terms of the "nearly 2 gigawatts" of credit enhancement for the unnamed second lab, which is the only piece of this Nvidia has quantified without naming.
- Whether supply stops binding. Every conclusion above rests on the queue being longer than the line. The moment Nvidia is demand-constrained rather than supply-constrained, concentration stops being an accounting question and becomes a revenue one.
Every figure attributed to Nvidia here — the "quarter of our business" line and its investment-grade qualifier, the approximately 70% fiscal-2028 growth guide, the nearly $50 billion of frontier-lab investments, the $108.5 billion of maximum guarantee exposure, the $500 billion of third-party capital and its six partners, the 12 gigawatts of OpenAI commitments and the $18/$25/$40 billion per-gigawatt series — is from the company's fiscal Q2 2027 results and earnings call, and is disclosed. Reported quarterly revenue and free cash flow are as Nvidia reported, from our June-quarter coverage of that print. Everything else is ours: the fiscal-2028 revenue range and the $167 billion quarter are arithmetic on the guide with an unguided fourth quarter held flat at the third quarter's level, the $216–480 billion span is the company's own per-gigawatt figures applied across its commitment book, and the four valuation rows are runs of our Nvidia model, whose rack price, shipment ceiling and growth rate are assumptions of ours rather than company disclosures.
